The Pursuit of Fairness in Artificial Intelligence Models: A Survey

Fairness in AI is complex because there's no single definition: The survey identifies that researchers use different definitions of fairness in current literature, meaning what counts as "fair" can vary depending on context, application, and stakeholder perspective—making it essential for developers to be explicit about which fairness criteria they're prioritizing.

Bias in AI systems can be both intentional and unintentional: The paper emphasizes that discriminatory outcomes affecting gender, ethnicity, and disabled people can occur without deliberate intent, highlighting that developers must actively work to identify and eliminate bias rather than assuming fairness happens by default.

Bias manifests differently across application domains: The survey documents cases of biased AI across sensitive areas like healthcare, education, and employment, suggesting that mitigation strategies need to be tailored to specific contexts rather than applying one-size-fits-all solutions.

Multiple technical approaches exist to address bias: Researchers have developed various techniques and approaches to mitigate bias in AI models, indicating that solutions are available—though the survey suggests practitioners need guidance on which approaches work best for their specific situations.

Fairness requires considering both ethics and user experience: The paper highlights that biased models impact how users experience AI systems and raises ethical considerations that should be contemplated during development and deployment, not just after problems emerge.